Variation in the Slope Coefficient of the Fama Regression for Testing Uncovered Interest Rate Parity: Evidence from Fixed and Time-varying Coefficient Approaches

We investigate the potential presence of time variation in the coefficients of the ''Fama regression'' for Uncovered Interest Rate Parity. We implement coefficient constancy tests, rolling regression techniques, and stochastic coefficient models based on state space modelling. Among six major US bilateral exchange rates we find significant evidence for stochastic time variation. Using the statistical equivalence between stochastically varying coefficients and conditional heteroscedasticity we derive a proxy for time-varying 'risk', and investigate whether it explains the well known "negative bias" or "foreign discount bias puzzle" in the foreign exchange rate literature. We contrast our identification scheme to the ARCH-in-mean approach for empirically identifying risk premia.